OpenAI has released two new AI models, and both of them are cheaper to run than the models they replace. On Monday, November 17, 2026, the company launched GPT-6.2 Astra, its new frontier model for the hardest professional work, and GPT-6.2 Sol, its efficient model for everyday tasks. Both are available now in ChatGPT Work, Codex and the OpenAI API, and both arrive with a clear message: more capability, fewer tokens, lower bills.
The launch matters because it is not a small tune-up. Astra cuts its token prices by 20 percent compared with GPT-6 Astra, while Sol holds the same low prices as GPT-6.1 Sol and still gets smarter. Both models deliver better results on demanding work while using fewer output tokens, so users can get more done inside the same plan limits.
What OpenAI announced on November 17
GPT-6.2 Astra is the flagship. OpenAI describes it as the model for your hardest projects: long documents, complex code, scientific work and professional deliverables that have to follow a template exactly. GPT-6.2 Sol is the everyday model, designed for the coding, writing and agent workflows most people run all day. OpenAI says many of its own employees already use Sol as their default for technical and non-technical tasks alike.
Both models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and in the API as gpt-6.2-astra and gpt-6.2-sol. Pro subscribers on the 500-dollar-a-month tier can also use Ultrafast, OpenAI’s highest-speed mode, with either model.
The timing is notable. Astra is the first upgrade to the line since GPT-6 Astra itself, because GPT-6.1 Astra was scrapped in September after it failed the company’s safety bar. The launch is also a direct challenge to Anthropic’s Claude Fable 5.2, which arrived on November 9, and to Google’s Gemini 4 family.
GPT-6.2 Astra: built for professional work
Astra’s headline claim is about quality on real professional jobs. OpenAI says its slides, spreadsheets and documents read less like AI output and more like the work of a person in the field, because the model follows the templates and conventions professionals actually use in law, sales, finance and accounting. The writing is more direct and specific, with fewer generic takeaways and less filler.
Slides get the same treatment. Astra is OpenAI’s best model yet at designing slides and following a template, even when it is given only a screenshot of a deck. It can match the layout, colours and typography down to the exact font size, and carry the template’s writing style into new slides. Early testers found the decks needed fewer edits.
A new high for scientists
OpenAI calls Astra the best model in the world for scientists, and it points to two benchmarks to make the case. On Terminal-Bench-Science 0.1, which tests how well models can carry out scientific work in a terminal, Astra scores nine percentage points higher than GPT-6 Astra at less than one-third of the cost. It also sets a new high on the Vals Science Index, which asks models to interpret real biological data and run experiments to infer how a simulated system works.
On OpenAI’s internal GDPval v2 benchmark of long-horizon work from real consulting, finance and design projects, Astra improves 6.2 percentage points over GPT-6 Astra while using 22 percent fewer output tokens. Sol improves 5.3 points over GPT-6.1 Sol.

GPT-6.2 Sol: the everyday workhorse
Sol is the model most people will actually use, and its story is about value rather than raw power. It costs the same as GPT-6.1 Sol in the API, yet OpenAI says it is smarter and more efficient, letting users complete more tasks on the same paid plan. It powers through multistep agent workflows, writes clearly and naturally, and is, in OpenAI’s words, intuitive to collaborate with.
The most striking number comes from Cognition’s FrontierCode 2.0, which tests whether changes to real codebases are good enough for maintainers to merge. On that test, Sol comes within half a point of Astra at about one-fifth of the cost, with both models running at maximum reasoning effort. On LifeVal, OpenAI’s test of challenging everyday tasks such as shortlisting kindergartens after a move or planning meals for a parent adjusting to a new diagnosis, Sol outscores the older GPT-6 Astra at about one-seventh of the cost.
Independent evaluator Artificial Analysis also published results on launch day. On its Intelligence Index, Sol scores 57, up from 55 for GPT-6.1 Sol, at a weighted cost of 1.01 dollars per index task. Astra scores 62, up from 57 for GPT-6 Astra, at 4.00 dollars per task. For comparison, Claude Fable 5.2 scores 65 but costs 26.54 dollars per task.
Sol keeps a context window of 1,050,000 tokens, and its knowledge cutoff moves from April 30, 2026 to June 1, 2026. Astra’s knowledge cutoff is May 18, 2026.
Prices: cheaper at the top, unchanged at the bottom
Astra’s standard API prices fall 20 percent from GPT-6 Astra: 8 dollars per million input tokens, 40 cents per million cached input tokens and 40 dollars per million output tokens. The cached-input rate carries a 60 percent discount. Sol keeps GPT-6.1 Sol’s rates exactly: 2 dollars per million input tokens, 10 cents per million cached input tokens and 10 dollars per million output tokens.

That makes Astra exactly four times Sol’s price per token in every matching category, which is why the choice between them is really a choice about the job. OpenAI recommends Sol for everyday coding and writing, and Astra for demanding work such as tracing a bug across services or producing a document that must follow a template precisely.
ChatGPT usage limits rise slightly too: OpenAI’s included-usage estimates go from 5 to 45 Astra messages per five hours to 6 to 50 on Plus and Standard Business plans, and from 15 to 160 to 17 to 180 for Sol.
How GPT-6.2 compares with Claude Fable 5.2
OpenAI picked its comparisons carefully, and readers should know these are the company’s own figures, not independent results. Still, the pattern is consistent across several tests, and it is the core of OpenAI’s pitch against Anthropic.
On Agents’ Last Exam, which tests multistep professional work, Astra scores 5.2 points higher than GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at just 5 percent of the cost. OpenAI notes, however, that Fable 5.2’s leaderboard results exclude nine tasks that used a fallback model, so the comparison is not perfectly clean.
On Harvey LAB-AA, which uses legal assignments and gives credit only when a deliverable meets every requirement with no material hallucinations, Astra more than doubles Fable 5.2’s pass rate at about one-sixth of the cost. A lawyer-specific configuration, Astra for Law, scores higher still.
The rivalry is familiar ground for gloznews.com readers. Google’s Gemini 4 Carbon, covered here last week, is the other model in this three-way race, and each launch now arrives with a benchmark chart aimed squarely at the other two.
Safety: the first “High” self-improvement rating
The launch also carries a safety milestone that OpenAI itself highlights. Astra is the first externally released model the company rates High for AI self-improvement, which measures a model’s ability to accelerate AI research. OpenAI says the designation reflects its assessment that the model’s impact is equivalent to giving every OpenAI researcher an additional, highly capable research engineer as an assistant, relative to a 2024 baseline.
Both models meet the Critical threshold for cybersecurity under OpenAI’s Preparedness Framework, and the High threshold for biological and chemical capabilities. On the company’s ExploitBench test, Astra reached a 53.5 percent mean arbitrary-code-execution rate at maximum reasoning, and Sol 50.5 percent, against 31.5 percent for GPT-6 Astra. The figures come from testing with safeguards disabled, and OpenAI is launching both models with tighter safeguards for higher-risk cybersecurity assistance.
How to get GPT-6.2 today
For most readers, the switch is automatic. Plus, Pro, Enterprise and Edu users will find Astra and Sol in the model picker in ChatGPT Work and Codex from today, and developers can call them in the API as gpt-6.2-astra and gpt-6.2-sol. Enterprise and Edu users may need their workspace administrator to switch the models on.
The sensible way to try them is to start with Sol. It costs a quarter of Astra per token, it is close to Astra on several coding tests, and it beats the previous generation’s flagship on everyday tasks at a fraction of the cost. Reach for Astra when the work is genuinely hard: a long document that must follow a template, a scientific analysis or a complex codebase change.
One caution: early reactions from users have been mixed, with some testers praising the speed and others calling the first builds rough around the edges. Test GPT-6.2 on real work before moving anything important onto it.
What happens next
The model race is now moving in weeks, not years. Google’s Gemini 4 family, Anthropic’s Claude Fable 5.2 and now OpenAI’s GPT-6.2 have all landed within days of each other, and each one undercuts the last on price while claiming the lead on benchmarks.
For now, the practical news is simple: ChatGPT and Codex users get two faster, cheaper models today, and developers get a frontier model at 20 percent below the price of its predecessor.
